Classification of Drowsiness Levels Based on a Deep Spatio-Temporal Convolutional Bidirectional LSTM Network Using...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Aviation Safety

Background:

• Human factors, including fatigue and drowsiness, contribute to over 70% of aviation accidents.
• Accurate monitoring of pilot mental states is critical for preventing accidents.
• Non-invasive brain-computer interfaces (BCI) show promise in recognizing mental states.

Purpose of the Study:

• To classify two mental states (alert and drowsy) and five detailed drowsiness levels from electroencephalogram (EEG) signals.
• To develop a novel deep learning model for precise drowsiness detection in pilots.
• To establish a new benchmark for detailed drowsiness classification using only EEG.

Main Methods:

• Acquired EEG data from ten pilots in a simulated night flight environment.
• Developed and applied a deep spatio-temporal convolutional bidirectional long short-term memory network (DSTCLN) model.
• Validated classification performance against Karolinska Sleepiness Scale (KSS) values.

Main Results:

• Achieved a grand-averaged classification accuracy of 0.87 (±0.01) for two mental states (alert/drowsy).
• Successfully classified five distinct drowsiness levels with a grand-averaged accuracy of 0.69 (±0.02).
• Demonstrated the first detailed classification of drowsiness levels using solely EEG signals.

Conclusions:

• The DSTCLN model effectively classifies pilot drowsiness levels from EEG signals.
• This approach shows high feasibility for real-time fatigue monitoring in aviation.
• Advanced BCI and deep learning offer significant potential for enhancing flight safety.